The human vision system is a very poorly understood mechanism that translates photons of various wavelengths into visual pictures that human brains can understand and respond to. The human vision system is extremely sensitive to differences in color.
When a color scanner or printer produces a color reproduction of an original image, even slight color variations produce images which do not look like they have the colors of the original image. The color variations could be in the paper (such as due to aging), the scanner light source, or the printer color toners or ink. For example, in scanners, the light source is generally a long cylindrical fluorescent bulb which is subject to a number of problems. When the light source is first turned on, the color balance of the light varies dynamically along the length of the bulb until the overall temperature stabilizes. During a long sequence of scans, the temperature of the bulb can change sufficiently to shift the color balance causing a color shift in the scanned images over time. Due to the need to frequently turn the light source on and off, each scan produces a slightly different color copy. Little research has been done into how to correct for the apparent visual color differences.
However, extensive research has been undertaken to predict a mathematical construct for visual images, such as photographs, called the White Point (WP). The WP is the illumination that occurred at the brightest point in the picture and represents what should be considered "white" in the final picture. It is assumed that every picture has some white objects or highlights in it. When the brightest object, with roughly equal amounts of red, green, and blue is found, the WP operation is constructed by determining the multipliers of the red, green, and blue parts of the brightest point so that the resulting red, green, and blue values will be made equal. Once this transformation is known for the brightest point in a picture, it is simultaneously applied to all the other points (which are called "dots") in the picture. The WP operation typically results in a final picture that looks much more realistic with respect to its original color balance.
There is a significant shortcoming of the simplistic WP operation described above. It is the assumption that there are some spectrally "white" objects in the picture. While this is true for the majority of pictures, there are also numerous cases where a spectrally "white" object is not present, for example, a close-up picture of a red barn with some blue and green metal signs attached to the barn's side. The dominant color would be red, and, in the reproduction process, this might be interpreted as a color cast problem. The brightest part of the picture would be the green signs. If the algorithm attempted to use the green area as the WP, then the resulting duplicate would be made very blue.
A solution is to measure the original image directly. In color photography, a more sophisticated type of "light meter" called a "photo spectroradiometer" is used. A photo spectroradiometer has to measure numerous points across the visual light spectrum and make a graph of the power at each wavelength that it has found. Once this graph is known, then an accurate representation of the original picture can be constructed by using colors duplicating the graph for different color variations in the paper, the scanner light source, or color toners or ink. In the example of the red barn with the blue and green signs, the photo spectroradiometer graph would show the proper proportions of red, blue, and green even on pale blue paper.
The problem is that a spectroradiometer is both big and expensive. A typical unit is 10 by 6 by 4 inches in size and costs between $5000 to $50,000 in 1998 dollars. It also requires a computer to read out its graphical data and apply it to the image in question. What is needed is a scanner or printer has a low-cost, small, portable spectroradiometer to indicate the color corrections necessary to compensate for the various variations inherent in the scanning or printing process.